Stable Diffusion 3作为稳定扩散模型的第三代版本,日前正式发布。新版本在图像渲染能力、多主题表达能力以及图像质量等多个维度实现重大提升,可生成效果与Midjourney不相上下,部分样本甚至质量更高。技术实现上,新版本模型参数规模范围为800MB至8GB,但具体技术细节尚未公开,仅透露采用了扩散型Transformer网络结构以及流匹配技术。相较此前版本,新版Stable Diffusion 3可望在更广泛的商业及研究应用中大放异彩。

Title: Stable Diffusion Image Generation Model Releases New Version
Keywords: image generation, model update, quality improvement
News content: Stable Diffusion 3, the third generation version of the Stable Diffusion model, was officially released recently. The new version achieves major improvements in image rendering capabilities, multi-theme expression capabilities, and image quality. The generated effects are comparable with Midjourney, and some samples even have higher quality. Technically, the parameter scale of the new version model ranges from 800MB to 8GB, but specific technical details have not been disclosed yet, only revealing the use of diffusion-type Transformer network structures and flow matching technology. Compared with previous versions, the new Stable Diffusion 3 is expected to shine in more extensive commercial and research applications.

【来源】https://www.qbitai.com/2024/02/122881.html

Views: 1

发表回复

您的电子邮箱地址不会被公开。 必填项已用 * 标注